biondizzle/DeepGEMM
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Cleanup some useless staffs

Browse Source
This commit is contained in:
Chenggang Zhao
2025-05-14 15:46:45 +08:00
parent ebf3d2f916
commit e2d6a107ef
3 changed files with 2 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
deep_gemm/include/deep_gemm/fp8_gemm.cuh
View File

@@ -36,7 +36,7 @@ template <uint32_t SHAPE_N, uint32_t SHAPE_K,
uint32_t kNumTMAMulticast, bool kIsTMAMulticastOnA,
GemmType kGemmType>
__global__ void __launch_bounds__(get_num_threads_per_sm<kNumTMAThreads, kNumMathThreadsPerGroup>(BLOCK_M), 1)
fp8_gemm_kernel(__nv_bfloat16* gmem_d, float* scales_b, int* grouped_layout,
fp8_gemm_kernel(float* scales_b, int* grouped_layout,
uint32_t shape_m,
const __grid_constant__ CUtensorMap tensor_map_a,
const __grid_constant__ CUtensorMap tensor_map_b,

2
deep_gemm/jit_kernels/runtime.py
View File

@@ -247,7 +247,6 @@ static void __instantiate_kernel() {{
config.hStream = stream
arg_values = (
gmem_d.data_ptr(),
scales_b.data_ptr(),
grouped_layout.data_ptr(),
shape_m,
@@ -257,7 +256,6 @@ static void __instantiate_kernel() {{
tensor_map_d,
)
arg_types = (
ctypes.c_void_p,
ctypes.c_void_p,
ctypes.c_void_p,
ctypes.c_uint32,